ASHS 2024 Conference

Advanced 3D Imaging for High Throughput Phenotyping of Horticultural Crops - Yu Jiang
Integrating UAV Imagery and AI to Forecast Vidalia Onion Yield and Quality - Marcelo Barbosa
Deep Learning Application for Field Phenotyping of Shoot Structure in Grapevine - Soichiro Nishiyama
Investigation of Using Hyperspectral Vegetation Indices to Assess Brassica Downy Mildew - Bo Liu
Effect of Innovative Laser Labeling Technology on Fresh Produce Quality and Safety - Manreet Bhullar
Cover Crop Decision Support Tools: Exploring the new suite of online cover crop tools - Esleyther Henriquez Inoa
CFD-based aerodynamic analysis under high wind velocity environment for multiple greenhouses - Anthony Kintu

Understanding plant growth and development is crucial for insights into plant structure and function, and recent advancements in AI-driven 3D imaging technologies have revolutionized the acquisition and analysis of high-fidelity plant models. These technologies enable accurate and rapid measurement of phenotypic traits, aiding breeders in developing new varieties and helping horticulturists optimize production management. The overarching goal of this study was to establish an AI-based 3D imaging and analysis pipeline specifically designed for detailed examination of horticultural crops at the organ level within controlled environments. We developed a robotic platform equipped with a rotating base and a high-resolution camera mounted on a robotic arm, allowing comprehensive imaging from any angle around the plant. Utilizing this robot, we generated 3D models of 30 hemp plants from two growth-rate categories in controlled environments, on a weekly basis. An AI model was developed to segment these 3D models into stems, branches, and leaves. Morphological traits were extracted from each category of the segmented organs, including stem length (i.e., plant height), stem diameter, branch length, branch diameter, leaf number, leaf area, and leaf aspect ratio. These measurements contributed to a classification model capable of distinguishing between fast and regular growth rates. Experimental results showed that the 3D imaging-derived measurements were highly correlated with human-derived measurements. In addition, the extracted traits were used as quantitative descriptors to classify hemp cultivars with different growth rates in CEA. Therefore, the developed pipeline can be used as an effective and efficient tool for breeding programs and CEA production management in the future.

Forecasting yield and quality of Vidalia onions allows the stakeholders to make decisions on the best time and place to harvest. While yield defines an important quantitative parameter, conversely, sweetness emerges as timely factor of quality. Traditionally, measuring these parameters requires a field team and routine laboratory for the assessments, making it a subjective, time-consuming, labor-intensive, costly, and not-scalable approach. However, image technology and artificial intelligence (AI)-based methods could improve decision-making strategies. In this study, we collected unmanned aerial vehicle (UAV) multispectral images of two Vidalia onions fields from crop establishment until the harvest date, totaling six sets of images for each field. Each flight was performed with approximately 15 days apart. At the harvest date, 50 samples were collected in each field to determine yield, while 10 samples were used for sweetness. To ensure the robustness of the dataset, both fields were combined into a single dataset. Consequently, we used machine learning (ML) algorithms to perform predictive models, namely multiple linear regression (MLR), random forest (RF), and support vector machine (SVM). The dataset was split into 70% and 30% for training and testing, respectively, and the predictions were performed using the test dataset. Regarding the assessment of the models, we used the metrics namely coefficient of determination (R2), mean absolute error (MAE), and root mean squared error (RMSE). The models with higher R2 and lower MAE and RMSE were the bests. Notably, the considerable correlation between yield and spectral data made the MLR model perform well as more complex models such as RF. Conversely, when there was a weak correlation between the sweetness and spectral data, RF model could perform much better. In short, both models (MLR, RF, and SVM) could perform well into a predictive model, which highlights the strength of spectral data for representing Vidalia onions either quantitative or qualitative parameters. Therefore, our study not only represents an innovation in the field of specialty crop production, but also brings ready-to-use solutions to improve the production process and introduce Vidalia onions into the concept of field technology.

In the cultivation of fruit trees and vines, plant architecture is a critical determinant of productivity. While there are considerable diversities in plant architecture, which can be modified through pruning in fruit production, a method for high-throughput measurement and recording of architecture has not yet been established, posing a limitation to research and development in this area. Here we evaluated Transformer-based architecture for detecting above-ground shoot network of grapevine in an outdoor vineyard condition. The problem here was defined as the detection of nodes (buds or branching points) and their physical relationships (internodes or edges) within plant images. We also developed an evaluation metric inspired by the inherent structure of plant shoots to efficiently smooth detected structures to more closely resemble realistic systems in plants. The proposed framework has been successfully applied to the detection task in outdoor condition with complex background. Through the application of this method, we have demonstrated that our proposed framework is capable of extracting topological parameters of dormant shoot architecture of grapevine that effectively models the shoot biomass in a large-scale vineyard.

Downy mildew, caused by Hyaloperonospora parasitica, poses a significant threat to Brassica oleracea crops, leading to substantial reductions in yield and marketability. This study seeks to assess various vegetation indices for detecting different levels of downy mildew infection in a Brassica variety, Mildis, using hyperspectral data. Through artificial inoculation with H. parasitica sporangia suspension, distinct levels of downy mildew disease were induced. Spectral measurements, ranging from 350 nm to 1050 nm, were performed on the leaves under controlled environmental conditions, and reflectance data were collected and processed. The Successive Projections Algorithm (SPA) and signal sensitivity calculations were employed to identify the most informative wavelengths, which were then used to develop Downy Mildew Indices (DMI). A total of 37 existing vegetation indices and three proposed DMIs were evaluated to assess downy mildew infection levels. The results revealed that a support vector machine achieved accuracies of 71.3%, 80.7%, and 85.3% in distinguishing healthy leaves from those with early (DM1), progressed (DM2), and severe (DM3) infections, respectively, using the proposed downy mildew index. The development of this novel downy mildew index has the potential to facilitate the creation of an automated monitoring system for downy mildew and aid in resistance profiling in Brassica breeding lines.

Introduction: Fresh produce is commonly associated with foodborne disease outbreaks and food recalls. To prevent the lethal impact of outbreaks, effective traceability is crucial. Produce items are traditionally labeled with price lookup (PLU) stickers. However, those stickers are environmental hazards, and frequent detachment of PLU stickers losses the information for traceability. Purpose: To investigate the effect of postharvest quality and microbial safety of laser labeling on fresh produce. Methods: Three horticultural crops, apple ‘Red Delicious ‘apple, cucumber, and green bell pepper, were procured from a local grocery store. Each produce was printed with a Quick Response (QR) code or text code using the laser engraver machine, followed by the application of edible wax. All produce was stored at 4° C temperature and 90% relative humidity during the study period. The postharvest quality was measured through fresh weight loss, QR code readability, and visual quality for 16 days. In another study, the laser-labeled produce was assessed for microbial contamination by quantifying artificially inoculated rifampicin-resistant E.coli (ATCC 25922) at an initial concentration of 106 CFU/mL. The experiment had five treatments: QR-coded labels followed by waxing or no wax, text-coded labels followed by waxing or no wax, and nontreated control. Results: Fresh weight loss for laser-printed produce was slightly higher than the nontreated control, but no difference in visual quality ratings was observed compared to the control. The population of rifampicin-resistant E.coli was statistically higher in all three produce labeled with text code compared to the nontreated control. However, QR-coded treatments were similar in the control. The application of wax did not facilitate microbial attachment. Significance: Laser labeling technology did not deteriorate the postharvest quality and susceptibility to microbial contamination. Hence this technology has the potential in commercial application as a better alternative to the PLU sticker to improve traceability.

Cover crop recommendations can be complex based on regional factors and different growing conditions. In order to combat these challenges, the Precision Sustainable Agriculture team (PSA) developed online tools that are readily available for producers to help them optimize cover crops on their operation. Tools include a species and variety selector tool, seeding rate calculator, nitrogen calculator, and economic decisions tool. These platforms look to help producers find success with cover crops that fit their operation’s needs.

In South Korea, approximately 65% of the land is mountainous or forested, which limits large-scale farming. Over 53,000 ha of land has been reclaimed from the sea and dedicated to the development of large-scale indoor agricultural complexes. Given the coastal climatic conditions and flat nature, these areas present unique challenges including stronger winds and colder temperatures compared to the inland, leading to high air velocities and operation costs in naturally ventilated greenhouses. Aerodynamic analysis is necessary to estimate crop risk factors and identify potential aerodynamic problems before the construction of these structures. Traditional studies have focused on using natural ventilation rates to estimate greenhouse suitability for plant growth. However, under scenarios of high wind velocity, this approach has critical limitations in accounting for crop damage resulting from high air velocity induced inside naturally ventilated facilities. This is tailored to the fact that ventilation efficiency in naturally ventilated structures increases with an increase in air velocity. High wind velocity induced inside greenhouses is associated with rapid CO2 depletion, stomatal dysfunction, leaf abrasion, mechanical stress etc., which critically affect crop yield and biomass development. Under high wind environment, crop damage resulting from high internal air velocities is an important factor that needs to be accounted for during design of indoor agricultural facilities. This study introduces a CFD model for designing greenhouse complex including multiple greenhouses and model analysis approach. We developed the Aerodynamic Crop Damage Index (ACDI), used it to analyze the model, and compared it to the convectional ventilation efficiency approach. The ACDI revealed a 2.2-fold variation in damage potential based on the greenhouse's location within the complex and a 15-fold variation attributable to wind direction, pinpointing significant damage risks in zones with the highest and lowest air velocities. In contrast, the convectional ventilation efficiency method only identified damage risks in low-velocity areas. ACDI has not only the potential to account for high air velocity effects in naturally ventilated greenhouses but also presents an opportunity for specialized greenhouse complex control and management according to greenhouse position and incoming wind direction. Future research should aim at refining the ACDI for better aerodynamic analysis in greenhouse complexes planning and its integration into greenhouse ventilation control systems.

Acknowledgments: This work was supported by “Regional Innovations Strategy (RIS)” through the National Research Foundation of Korea (NRF) funded by Ministry of Education (MOE) (2024RIS-008)

This year Conviron is launching three new products:
•             GEN1000-ECO (introduction date: April 16, 2024)
•             ConvironDirect (introduction date: March 4, 2024)
•             PGR15/E15 LED Retrofit (introduction date: Jan 5, 2024)
GEN-1000-ECO:
The GEN1000-ECO is a new compact reach in chamber ideal for short and tall plant research that comes standard with humidity control and energy efficient features such as a smaller compressor and LED lighting - for up to 15% reduced energy consumption. Low, medium and high light options are available to meet a range of research requirements.
ConvironDirect:
ConvironDirect is a new premium software tool that enables users to manage chamber setpoints and actual conditions remotely through any building LAN connected desktop, notebook or handheld mobile device. ConvironDirect is ideal for users that have Conviron reach-in plant growth chambers or walk-in rooms and want a seamless connection to their chamber, their plants, and their data from virtually anywhere.
PGR15/E15 LED Retrofit:
Fluorescent lamps such as T5, T8 and T12 have been the standard for many years and have been used in tens of thousands of plant growth chambers around the world. However, fluorescent lighting is trending towards obsolescence and replacement lights are increasingly difficult to source economically. Conviron is now offering a retrofit for aged PGR15-E15 chambers to enable users to take advantage of the latest LED lighting technology and save up to 80% on energy costs.

Introduction and Overview

Speaker: Kathryn Orvis
Professor
Department of Horticulture and Landscape Architecture
Purdue University
625 Ag Mall Drive
West Lafayette, IN 47907-2010

Title: Digital Agriculture and AI on Specialty Crops Production

Description: Digital agriculture is the 4th agricultural revolution and Artificial Intelligence (AI) is part of it. Currently, in the "connected agriculture"; era, many technologies have been released on the marked regarding the use of multispectral
sensors for many purposes in agriculture. This talk is going to cover information on how to use Digital Agriculture online platforms to process multispectral imagery, and how AI can be used to collect individual in-field plant data.

Speaker: Luan Pereira de Oliveira
Assistant Professor and Precision Agriculture Extension Specialist
Department of Horticulture
University of Georgia
139 Engineering Building
2329 Rainwater Road
Tifton, GA 31793

Title: Bringing the Future of AI to the Farm.
Description: In this talk, we will cover the multitude of use cases where AI can be applied in farming – from weed detection and robotics to Generative AI-based farm assistants and Virtual Reality. We go through the industry trends of applied Artificial Intelligence and think big about farm automation for the future.

Speaker: Justin Hoffman
Chief Technology Officer of AgTechLogic


Title: From Concept to Impact: The Evolution of Moss Robotics through Industry-
University Collaboration

Description: Moss Robotics' journey began with a project focused on autonomous driving technology for tree nurseries, born out of a collaboration between Carnegie Mellon University, Robotics Institute and Hale; Hine Nursery in Tennessee. In this talk, we share the story of how we discovered the real value our solution could offer to growers, and how we refined our ideas through continuous iteration. This process transformed moss robotics from a simple concept into the company it is today. We will cover the steps of our evolution, emphasizing the practical benefits of combining academic research with industry needs to innovate effectively. Additionally, we look ahead to how emerging technologies might further influence our growth and the agricultural industry as a whole, aiming for advancements in farming practices that are both technologically sophisticated and grounded in real-world applications.

Speaker: Di Hu
Founder and CEO
Moss Robotics

Title: AI-Enhanced Computer Vision for Crop Monitoring in Controlled Environment
Agriculture

Description: Controlled environment agriculture (CEA) production remains expensive due to high operation costs. Growers can reduce production costs by nurturing crops with data, however, the data is highly diverse, and growers lack the expertise to analyze this data to derive actionable insights for informed decision-making. In addition, traditional crop monitoring is carried out manually, which makes it unfeasible to collect data daily to get actionable insights for high yields. Recent advancements in sensing and computing technologies, such as AI, computer vision, edge computing, and edge-
cloud integration, have opened opportunities to develop data-driven technologies to enhance decision-making capabilities. Integrating AI and computer vision technologies has emerged as a transformative toolset that can collect real-time plant data at high spatial and temporal resolutions, pivotal in optimizing resource management and maximizing production. The CE Engineering lab delves into cutting-edge computer vision applications within CEA, focusing on various applications, including phenotyping leafy greens, yield estimation, disease monitoring, and plant spacing optimization. This presentation will explore the details of lettuce phenotyping, disease classification, strawberry fruit classification, and yield estimation. We will delve into the technical aspects of these algorithms, including image processing techniques, machine learning models, and data integration strategies. This presentation will showcase state-of-the-art deep learning approaches, including segmentation algorithms, model training, and deep classifiers. Overall, this presentation aims to provide insights into the transformative potential of computer vision in CEA, offering a glimpse into the future of data-driven and sustainable CE production.

Speaker: Azlan Zahid
Assistant Professor,
Department of Biological and Agricultural Engineering
Texas A&M AgriLife Research
Texas A&M University System
Dallas, TX 75252, USA


Panel: 30-minute panel with the above speakers, to allow time for Q&A and discussion.